Dataset bounding box vertices are incorrect

[MCSCC]

Hello.
I have an issue with my dataset, for training my AutoML model. When selecting bboxes on an image the following error is showing.

Error message: The specified bounding box vertices are incorrect at: , , , , , , , , , , , , , , . Please correct the coordinates such that they adhere to the BOUNDING_BOX column specification.

It was fine just 2 days ago, and it started showing this message yesterday on both new and old images on the dataset. 

PS: The old images in the dataset were fine and now it shows the error message with null coordinates. 

[ibaui]

Hi @MCSCC ,

Welcome to Google Cloud Community!

It seems you're encountering a specific error message when selecting bounding boxes in your dataset for AutoML. The error message "The specified bounding box vertices are incorrect" indicates a problem with the format or values of your bounding box coordinates in the BOUNDING_BOX column. Here are some potential reasons and suggestions you might consider to address the issue: 

Incorrect Coordinate Format:

• Check the Format: Your AutoML system likely expects bounding box coordinates in a specific format (e.g., xmin, ymin, xmax, ymax or x_center, y_center, width, height, possibly normalized to [0, 1]). With this, ensure that the coordinates are in the correct format.
• Verify Data Types: Ensure that your bounding box coordinates are of the correct data type (e.g., integers, floats). A mismatch could cause errors.

Data Corruption:

• Inspect the Dataset: Manually examine the affected images and their corresponding bounding box coordinates to identify any anomalies.
• Re-annotate: If necessary, re-annotate the images with correct bounding boxes.
• Data Conversion: If you recently converted your dataset ot performed any transformations on it, double check the conversion process.

Issues with the AutoML Tool:

• Version Updates: Did you recently update your AutoML platform, its libraries, or any related software? Updates can sometimes introduce changes in the expected data format or how it handles bounding boxes
• Configuration: Review the configuration settings for the AutoML tool, especially those related to the bounding box input.

Specific to Your Dataset:

• Null Coordinates: You mentioned the error message shows “null coordinates” for old images. This indicates that the bounding box coordinates might have been accidentally removed or overwritten, might be missing or not properly defined. Make sure the coordinates are within the image dimensions and are not null or empty.
• Image Transformations: If you have applied any image transformation (resize, cropping) to your images, ensure that the bounding box coordinates have been adjusted accordingly.

You can refer to this documentation for more details regarding the requirements for datasets used to train AutoML models.

If the issue persists, I suggest contacting Google Cloud Support as they can provide more insights and take a deeper look at your issue to see if the behavior you've encountered is a known issue or specific to your project.

I hope the above information is helpful.

[MCSCC]

Hello,

Thank you for the reply. 

I do not think it has anything to with with the following: Incorrect Coordinate Format, Data Corruption, Null Coordinates or Image Transformations. 

I have experimented and checked my data multiple times, and the error still persists. Then I have tried creating a new dataset and uploaded few images to check. To my surprise, the error showed up even in the new dataset. 
What can I do to report this issue? I think it has something to do with Google Cloud Vertex AI dataset Itself. 
Again thank you for you reply